Machine Learning | e-Portfolio

In his article, Schwab shares his views on a 4th technological revolution and how it will change and redefine our daily life at its core.
Although this drastic change will disrupt and affect every industry worldwide it could potentially yield positive effects on businesses and our economy; However, he warns that it could also create polarised categories amongst the work force. Schwab also presses for governments to emphasise focusing on security and promoting transparency to enhance communication with the public.To summarise, Schwab predicts an unprecedented transformation on our society, which if we intend to manage successfully, will require all parties to be involved.

This underlying dependency to technology can be illustrated by the consequences resulting from disrupted IT operations in the retail industry.

Amazon Web Services is a leading online infrastructure provider which is used and relied upon by 2.38 million businesses worldwide and, it owns 50.1% of the cloud platform market. On the 7th of December 2021, Amazon experienced a major outage, suspending multiple services, causing unimaginable repercussions. It affected colleges causing the postponement of examinations, household appliances seized to function, and online deliveries were cancelled. The outage had a domino effect on all Amazon’s retailers as well as affecting major media players such as Disney, PlayStation Network, Slack, Netflix and Snapchat.

The scale of this disruption demonstrates how brittle our economy is, how tightly coupled it has become with digital technology and can no longer perform without it.
Because of interweaving technical intricacies and scale, attempting to prevent this sort of incident and expecting systems to run with absolute resiliency seems nearly impossible. Could we expect governments and lawmakers to enforce frameworks to emphasise private providers’ accountability in order to guarantee compensation for loss incurred by such events.

References:

The Fourth Industrial Revolution: what it means, how to respond (By Klaus Schwab)
Dead Roombas, stranded packages and delayed exams: How the AWS outage wreaked havoc across the U.S. (By Annie Palmer Published Thu, Dec 9 2021)
View HG Insights’ full market report: The AWS Ecosystem in 2024
AWS Outage Analysis: December 7 & 10, 2021 (By Internet Research Team | December 7, 2021)

Gavin mentions the Knight Capital software bug and the catastrophic financial impact and loss that this technical negligence yielded. This was a result of human errors dating back to 2003 when an engineer at Knight Capital had left deprecated server code in their SMARS application service and nobody had noticed this oversight. During a refactor 2 years later, the tests for power peg were breaking, so they were deleted. Nobody was using the long-deprecated option, so there was no longer a need to check its correctness. In 2012 when a new feature flag needed to be added to SMARS and an engineer reused the deprecated flag where no test failed during deployment since the old test had been removed.

As a result, the argument for the exponential speed in which technology and machines are evolving highlights the question whether we can keep up with technological progress and if we can, what measures are we willing to take.

References:
The Fourth Industrial Revolution: what it means and how to respond | World Economic Forum.
Deploy Gone Wrong: The Knight Capital Story | by Alex Ponomarev | Engineering Manager’s Journal | Medium
The Knight Capital Disaster - Speculative Branches

Ben shows the precariousness our society exposes itself to by relying exclusively on interconnected information systems. To convey and illustrate his argument, Ben has chosen the CrowdStrike incident which occurred in July 2024 demonstrating the chaos which affected various sectors and industries and was caused by a failed security patch update, allowing attackers to infiltrate their target network. The first critical vulnerability is that the attackers exploited outdated or not appropriately secured servers (Naseer, 2024)

Data governance and security are determining paradigms to this underlying and ubiquitous digital layer which is allowing our society to function. Proactively addressing AI-based security issues is a key factor for an industrial environment with smart factories, autonomous systems, CPS, IoT, cloud computing, and big data (de Azambuja et al., 2023). Organising actors to proactively protect against those threats is exemplified by the importance and the scope of cyber warfare.

Ben raises concern in delegating cybersecurity tasks to AI and machine learning. A recent study revealed that organisations are increasing the pace of adoption of AI/ML in cybersecurity and overall, close to three-quarters of firms surveyed admitted that they were testing use cases for AI/ML for cybersecurity (Kinyua and Awuah, 2021).

In handing over key responsibilities such as decision making and deployment to machine learning algorithms, Ben also points at the risk of reducing human intellectual skills which is a recurrent concern associated with the rise of artificial intelligence. (Ahmad et al., 2023)

References:
Ahmad, S.F. et al. (2023) ‘Impact of artificial intelligence on human loss in decision-making, laziness and safety in education’, Humanities & social sciences communications, 10(1), p. 311. de Azambuja, A.J.G. et al. (2023) ‘Artificial intelligence-based cybersecurity in the context of Industry 4.0—A survey’, Electronics ETF [Preprint]. Available at: https://doi.org/10.3390/electronics12081920. Kinyua, J. and Awuah, L. (2021) ‘AI/ML in security orchestration, automation and response: Future research directions’, Intelligent automation & soft computing, 28(2), pp. 527–545. Naseer, I. (2024) ‘The crowdstrike incident: Analysis and unveiling the intricacies of modern cybersecurity breaches’, World Journal of Advanced Engineering Technology and Sciences, 13(1), pp. 728–733.

Many thanks for your responses, of which both are insightful. In reference to Schwab’s 2016 paper on the Fourth Industrial Revolution, we can take it as an outset to the discussion of the effects of modern day technology which is evolving at a rate that can be difficult to keep up with.

However, as we know, it is imperative that we maintain the momentum of this rapid growth and change in order to provide a supportive and safe environment for global business, technology and finance to flourish; a hybrid cloud architecture sounds like a very good compromise to supply security in both the private and public cloud but one guarantees absolute privacy and the other, easily accessible information. It is interesting to see in the Nguyen – Sondano table, that proper training is the top concern for mitigating cloud security risks, followed closely by the implementation of IPS.

References:
Khan, S.U. and Ullah, N. (2016) ‘Challenges in the adoption of hybrid cloud: an exploratory study using systematic literature review’, The Journal of Engineering, 2016(5), pp. 107–118. Nguyen, D.S. and Sondano, J. (2023) ‘Resilience and Stability in Organizations Employing Cloud Computing in the Financial Services Industry’, Journal of Computer and Communications, 11(4), pp. 103–148. Website (2016). Available at: https://www.weforum.org/stories/2016/01/the-fourth-industrial-revolution-what-it-means-and-how-to-respond/.

            import numpy as np


            # Jack	Y	N	P	N	N	A
            # Mary	Y	N	P	A	P	N
            # Jim	  Y	P	N	N	N	A

            #J(A, B) = (A & B) / (A | B)

            def jaccard_coef(x,y):
            intersection = np.logical_and(x, y)
            union = np.logical_or(x, y)
            similarity = intersection.sum() / float(union.sum())
            return similarity



            def binary_encode(s):
            bin = {"Y": 1, "P": 1, "N": 0, "A": 0}
            encoded_list = []
            for k in s:
            encoded_list.append(bin.get(k))
            return encoded_list


            dataset = {
            "Jack": ["Y", "N", "P", "N", "N", "A"],
            "Mary": ["Y", "N", "P", "A", "P", "N"],
            "Jim": ["Y", "P", "N", "N", "N", "A"]
            }

            jack = binary_encode(dataset["Jack"])
            mary = binary_encode(dataset["Mary"])
            jim = binary_encode(dataset["Jim"])

            print(f'Jack: {jack}')
            print(f'Mary: {mary}')
            print(f'Jim: {jim}')

            print('Jack & Mary: ',jaccard_coef(jack, mary))
            print('Jack & Jim: ', jaccard_coef(jack, jim))
            print('Jim & Mary: ', jaccard_coef(jim, mary))
        

            Output:

            Jack: [1, 0, 1, 0, 0, 0]
            Mary: [1, 0, 1, 0, 1, 0]
            Jim: [1, 1, 0, 0, 0, 0]

            Jack & Mary:  0.6666666666666666
            Jack & Jim:  0.3333333333333333
            Jim & Mary:  0.25
        

        import numpy as np

        def gradient_descent(x,y):
            m_curr = b_curr = 0
            iterations = 50
            n = len(x)
            learning_rate = 0.075

            for i in range(iterations):
                y_predicted = m_curr * x + b_curr  #y = mx + b
                cost = (1/n) * sum([val**2 for val in (y-y_predicted)])
                md = -(2/n)*sum(x*(y-y_predicted))
                bd = -(2/n)*sum(y-y_predicted)
                m_curr = m_curr - learning_rate * md
                b_curr = b_curr - learning_rate * bd
                print ("m {}, b {}, cost {} iteration {}".format(m_curr,b_curr,cost, i))

        x = np.array([1,2,3,4,5])
        y = np.array([5,7,9,11,13])

        gradient_descent(x,y)
        

        Output
        --------
        m 4.6499999999999995, b 1.3499999999999999, cost 89.0 iteration 0
        m 1.0200000000000005, b 0.405, cost 53.734999999999985 iteration 1
        m 3.8047499999999994, b 1.2352499999999997, cost 32.557024999999975 iteration 2
        ...
        m 2.178758038970076, b 2.354664382435115, cost 0.07979893231689464 iteration 48
        m 2.174208302573649, b 2.371023607533314, cost 0.07580328391789995 iteration 49
        

In “The Language Machines”, Matthew Hutson undertakes a critical analysis of Large Language Models technological advances and the potential societal and ethical risks associated with their use.
The author discusses the outstanding language fluency of an LLM but he describes the characteristics of its process as purely mathematical. It works by observing the statistical relationships between the words and phrases it reads, but doesn’t understand their meaning (A remarkable AI can write like humans — but with no understanding of what it’s saying. By Matthew Hutson, 2021).

The lack of reasoning or understanding labeled as “nonsensical answer” which he observed in the results yields by prompting such LLMs has raised concerns amongst the AI community describing the models as “stochastic parrots” (Bender et al., 2021).
This inequality is exemplified as an inherent bias named the “coded gaze” by Dr Buolamwini. As a student, she reported a facial recognition software failure while working on a university project. After researching the cause of the glitch, she deduced that companies had introduced bias in their models by training their neural networks with uniform data. As a result, the lack of diversity in the data rarely exposed the models to similar complexion and the software wasn’t able to detect her face. (Buolamwini, 2023)

The benefits in using artificial intelligence in healthcare has proven to be a logical course to take globally, particularly for the NHS. The UK Government is investing in the National AI Strategy, investing money into research for the sake of progress and cost effective budgeting. The introduction of a smart stethoscope has been used to identify patients with suspected heart disease, the results proved to be accurate in its detection of heart issues; the results are promising so far, giving a strong indication leaning favourably for GPs to use the stethoscope in order to spot disease well in advance, in turn not only saving time but money spent on referrals, thus improving diagnosis time for the patient as well. Artificial intelligence has not only shown its capabilities in diagnosing lung cancer earlier, (as opposed to using the Brock score), but it has shown that it can detect early signs of other conditions and disease.(Artificial intelligence: 10 promising interventions for healthcare, 2023)

References:
A remarkable AI can write like humans — but with no understanding of what it’s saying. By Matthew Hutson (2021) in. Cham, Switzerland: Springer International Publishing.
Artificial intelligence: 10 promising interventions for healthcare (2023). NIHR Evidence. Available at: https://doi.org/10.3310/nihrevidence_59502.
Bender, E.M. et al. (2021) ‘On the dangers of stochastic parrots: Can language models be too big?’, in Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency. FAccT ’21: 2021 ACM Conference on Fairness, Accountability, and Transparency, New York, NY, USA: ACM. Available at: https://doi.org/10.1145/3442188.3445922.
Buolamwini, J. (2023) Unmasking AI: My Mission to Protect What Is Human in a World of Machines. Random House.

In his post, James highlights the opportunities and challenges posed by the use of LLMs and more specifically within the healthcare industry where their transformative impact suggests an over reliance of their capacity.

Large Language Models have proven to be beneficial in many areas of a rapidly growing revolution in a new era of machine learning. Matthew Hutson conveys a doubtful opinion in his article “The Language Machine”; there are the potential risk factors in LLMs used by businesses and organisations in relying on what is still an emerging technology, because it is easy to access information and scripts, reducing time and money spent on researching and composing the literature manually it is easy to get used to relying on it, when it can be less than perfect.

Once again, the associated risks and ethical risks of becoming reliant on AI chatbots such as ChatGPT are obvious when clinicians, although trained and more than capable, use GPT4 for accessing and writing medical research papers. This, in the healthcare system, although beneficial, can also have its drawbacks.

In the research paper “The Global Landscape of AI Ethics”, its findings reflect the concerns of many people who, although embrace the fifth revolution it reveals the five principal concerns in AI:- transparency, justice and fairness, non-maleficence, responsibility and privacy. The landscape is wide and with time and prudence, on our part, LLMs can work well with us as opposed to “for us”.

References:
Li, F., Ruijs, N. and Lu, Y. (2022) ‘Ethics & AI: A Systematic Review on Ethical Concerns and Related Strategies for Designing with AI in Healthcare’, AI, 4(1), pp. 28–53.
Website (2019). Available at: https://www.researchgate.net/publication/335579286_The_global_landscape_of_AI_ethi cs_guidelines.

In these times of using AI, it is proving to be a powerful writing tool for organisations from businesses to healthcare, and proving to be a valuable instrument in research, writing papers, composing emails and completing administrative jobs whereby that time can be implemented for expanding on other factors in an efficient capacity. However, as with everything that is created for making life easier and cost effective it comes at a cost.

The creative industry is under threat of becoming the worst hit and as such, companies that are using AI for writing be it creative or content, images, and artwork, the author and the artist are unable to compete against the speed and efficiency of AI generated pieces. However, the generative tools of AI are at risk of running out of ideas that would usually be taken from real authors and artists because many in the creative sector are reducing their pieces as a negative response to an infringement of their work; in turn this would mean that businesses who have previously enjoyed the ease of accessing creative ideas, will run the risk of struggling without human artists. So, there is the ethical side to revolutionary progress, which if abused can cause problems in the near future when humans revolt against the use of their craft being snatched away. Of course, the concern for job displacement for human workers from the labour force to office work and the medical sector is very prevalent across the globe. However, AI is far from perfect, and Matthew Hutson encapsulates the current situation by simply stating that “What we have today is essentially a mouth without a brain”.

References:
Blogs (2023) Sogeti Labs. Available at: https://labs.sogeti.com/ (Accessed: 27 January 2025). De Cremer, D., Bianzino, N.M. and Falk, B. (2023) How Generative AI Could Disrupt Creative Work, Harvard Business Review. Available at: https://hbr.org/2023/04/how-generative-ai-could-disrupt-creative-work (Accessed: 27 January 2025). A remarkable AI can write like humans — but with no understanding of what it’s saying. By Matthew Hutson (2021) in. Cham, Switzerland: Springer International Publishing.

The use of LLMs in policing and surveillance activity has shown how predictive models can amplify the marginalisation of certain communities. The data selection process used to train these algorithms needs to be regulated and scrutinised so that we can avoid societal inequality(Jain, 2018).

Large Language Models in healthcare can pose a beneficial means in assisting doctors, thus saving time leading to prompt diagnoses and early intervention for patients. Transitioning to digital imaging in the medical industry has leveraged the use of CNNs for patterns recognition and classifying images. Exciting gains have also been made in predictive biomarker analysis. ML models have been used to identify predictive signatures from peripheral blood samples and tumour biopsy material, including analyses of whole-genome profiles(Hunter, Hindocha and Lee, 2022)

However we cannot ignore the “robotic responses” conveyed from LLMs based systems. Although the accuracy of CNNs in the medical field is progressing at an amazing rate helping with prognosis, diagnosis and assisting in tailoring patients treatment the necessary bond between patients and their doctor remains paramount. In diagnosing patients, human doctors have a unique way of making behavioral observations and showing a sense of empathy with patients that no machine can.(Cordero, 2023)

Collaborative AI in an ideal world, would yield the dream team; LLMs can work on the time-consuming tedious tasks that we as humans prefer to pass on to others. We are able to work on projects that computers cannot and in turn productivity higher and conducive to the increase of work generated between humans and artificial intelligence.(Slack, no date)

References:
Cordero, D., Jr (2023) ‘The downsides of artificial intelligence in healthcare’, The Korean Journal of Pain, 37(1), p. 87.
Hunter, B., Hindocha, S. and Lee, R. (2022) ‘The role of artificial intelligence in early cancer diagnosis’, Cancers, 14. Available at: https://doi.org/10.3390/cancers14061524.
Jain, A. (2018) ‘Book review: Weapons of Math Destruction by Cathy O’Neill’, Economics of Networks eJournal [Preprint]. Available at: https://doi.org/10.2139/ssrn.3187660.
Slack (2024) Collaborative Intelligence: People and AI Working Smarter Together, Slack. Available at: https://slack.com/blog/collaboration/collaborative-intelligence-people-and-ai-working-smarter-together.